- Email: [email protected]
In vivo two-photon imaging analysis of development of direction and orientation selectivity in the mouse visual cortex
Abstracts / Neuroscience Research 71S (2011) e108–e415
e257
P3-j19 Response properties of V2 neurons to combination of two local spectral components
P3-j21 Posterior parietal neurons represent the perceived pursuit direction in the Duncker illusion
Takayuki Nakazono 1 , Minami Ito 2,3 , Kunihiro Asakawa 3,4 , Izumi Ohzawa 1
Naoko Inaba 1,2 , Michael E. Goldberg 1
1
Grad. Sch. of Frontier Bioscie, Osaka Univ., Osaka, Japan 2 National Institute for Physiological Sciences, Okazaki, Japan 3 The Graduate University for Advanced Studies, Okazaki, Japan 4 The Jikei University School of Medicine, Tokyo, Japan Previous studies have shown that many neurons in V2 and V4 respond better to simple shapes such as angles or curves than to gratings which are the best shape for neurons in V1. These studies, however, used stimulus sets defined in a parameter space selected by the experimenters. These stimuli are limited in that selectivities to other stimulus parameters and configurations cannot be explored by definition. One way to sidestep this issue is to use random stimulus sets. Here we used dynamic random noise as stimulus set and recorded neural responses of neurons in V2 in anesthetized monkey. Since all high-order shape selectivities of neurons emerge from combinations of output of multiple V1 neurons, we calculated response properties to combinations of two local spectral components, each defined at (x, y, fx, fy), where (x, y) is position and (fx, fy) is spatial frequency/orientation. Response magnitude was defined by the product of the amplitudes of the two spectral components. Spectral combinations were examined exhaustively within the sampled frequency domain. We recorded 416 neurons from monkey V2. Half of neurons showed interactions to a combination of two spatially adjacent spectral components, but these response maps were somewhat noisy and Zscores were not so high (<7). On the other hand, many neurons showed the strongest interaction to the combination of two spatially overlapped spectral components with large Z-scores (>7) (N = 149). These results suggest that neurons in V2 code combinations of two spectral components. Research fund: KAKENHI 22135006, 22300110(IO), 18300111 (MI) and Global COE.
1 2
Department of Neuroscience, Columbia University Medical Center, NY, USA Japan Society for the Promotion of Science (JSPS), Tokyo, Japan
The world around us appears to remain stationary even when our eyes are in motion. This perceptual stability is caused by an internal estimation of the external visual world derived from both retinal and extra-retinal signals. However, the human visual system frequently misinterprets the meaning of retinal motion. Karl Duncker (1929) showed that when a stationary object on the retina is embedded in retinal background motion, the subject perceives the object to be moving in a direction opposite from that of the background. When a human or monkey pursues a target moving orthogonally to a flow field, although the eye movement is accurate, the eye movement and target motion direction are perceived to be moving in a diagonal direction away from the flow field. When humans or monkeys make memory-guided saccades to targets flashed at the beginning of an epoch of pursuit across an orthogonal flow field, their eye movements are inaccurate because they compensate for the pursuit movement the subjects thought they made but actually did not. To understand the neuronal mechanisms underlying this illusion we investigated the neuronal responses of pursuit-related neurons in area MST while the monkeys pursued a target moving across an orthogonal flow field and then made a memory-guided saccade to a stimulus which flashed at the beginning of the pursuit epoch. We asked whether the directional selectivity of each pursuit cell in area MST represents the actual direction of the pursuit trajectory or the perceived direction of ongoing pursuit as measured by the error in the memory-guided saccade. We found that the direction tuning profiles of a significant number of MST pursuit cells represented the direction of the perceived pursuit trajectory, not the pursuit eye movement that monkey actually performed. These results suggest that area MST compensates for self-induced motion by using the perceived direction of pursuit, rather than a corollary discharge of the eye movement itself.
doi:10.1016/j.neures.2011.07.1119 doi:10.1016/j.neures.2011.07.1121
P3-j20 From multi-neuron recordings to a brain-based machine: A method to extract and utilize nonlinear population coding manners for constructing artificial visual recognition system Hiroki Kurashige , Hideyuki Cateau RIKEN BSI-TOYOTA Collaboration Center, RIKEN, Wako, Japan We propose a method to construct a brain-based visual recognition system based on the in vivo neural activity data recorded from multiple neurons in the visual cortex of an animal watching visual stimuli. We demonstrate that our method works for a data taken from simulation. Our method extracts visual features represented well by the neural activity of an animal using the so-called kernel canonical correlation analysis (KCCA) which is a multivariate statistical method to detect nonlinearly correlated pairs of visual features and neural activity patterns, and incorporates them into artificial visual recognition systems. The extracted visual features are considered useful for visual recognition because the features are probably used in the brain performing sophisticated visual recognition. We represent the visual features in terms of a kernel function, that we call NC kernel, of kernel method which is good formalism to incorporate the features to artificial recognition systems. Attaching the NC kernel to a support vector machine (SVM), which is frequently used artificial recognition system, we construct a brain-based SVM incorporating the visual features. With the data obtained from the simulation of artificial neural network (ANN) instead of real brain, the “ANN”-based SVM performed the visual discrimination task requiring the knowledge which the ANN has. Our SVM correctly performed the task while usual SVM did not. This result suggests the successful transfer of the visual features from the ANN to the SVM. Moreover, we show that the NC kernel works well even if the visual stimuli presented to the ANN are non-specific to the task for the SVM. This is desired nature because we generally cannot specify the visual stimuli required by the task in the realistic application such as car-mounted camera. We finally discuss the possibility that we can use natural images as task-nonspecific visual stimuli to construct an effective NC kernel in realistic situations. doi:10.1016/j.neures.2011.07.1120
P3-k01 In vivo two-photon imaging analysis of development of direction and orientation selectivity in the mouse visual cortex Madoka Narushima 1 , Nathalie L. Rochefort 2 , Christine Grienberger 2 , Nima Marandi 2 , Arthur Konnerth 2 1
Dept. Physiol., Sch. of Med., Tokyo Women’s Medical Univ., Tokyo, Japan Inst. fuer Neurowissenschaften, Technische Univ. Muenchen, Munich, Germany 2
Previous studies of the ferret visual cortex indicate that the development of direction selectivity requires visual experience. In spite of the usefulness of rodents as a model system for studying mechanisms of functional development of neuronal circuits, it has been less understood that how the rodent visual system develops. Here, we used two-photon calcium imaging to study the development of direction-selectivity in layer 2/3 neurons of the mouse visual cortex in vivo. The number of visual responsive cells gradually increased after eye-opening. Surprisingly, at eye-opening nearly all orientation-selective neurons were also direction-selective. During later development, the total number of neurons responding to drifting gratings increased in parallel with the fraction of neurons that were orientation—but not direction-selective. Our experiments demonstrate that direction-selectivity develops normally in dark-reared mice, indicating that the early development of direction-selectivity is independent of visual experience. Furthermore, remarkable functional similarities exist between the development of direction-selectivity in cortical neurons and the previously reported development of direction-selectivity in the mouse retina. Together, these findings suggest a new experience-independent circuit mechanism for the development of direction-selectivity in the mammalian brain that is distinctly different from that previously found in ferrets. Research fund: JSPS Postdoctoral Fellowship for Research Abroad. doi:10.1016/j.neures.2011.07.1122
e257
P3-j19 Response properties of V2 neurons to combination of two local spectral components
P3-j21 Posterior parietal neurons represent the perceived pursuit direction in the Duncker illusion
Takayuki Nakazono 1 , Minami Ito 2,3 , Kunihiro Asakawa 3,4 , Izumi Ohzawa 1
Naoko Inaba 1,2 , Michael E. Goldberg 1
1
Grad. Sch. of Frontier Bioscie, Osaka Univ., Osaka, Japan 2 National Institute for Physiological Sciences, Okazaki, Japan 3 The Graduate University for Advanced Studies, Okazaki, Japan 4 The Jikei University School of Medicine, Tokyo, Japan Previous studies have shown that many neurons in V2 and V4 respond better to simple shapes such as angles or curves than to gratings which are the best shape for neurons in V1. These studies, however, used stimulus sets defined in a parameter space selected by the experimenters. These stimuli are limited in that selectivities to other stimulus parameters and configurations cannot be explored by definition. One way to sidestep this issue is to use random stimulus sets. Here we used dynamic random noise as stimulus set and recorded neural responses of neurons in V2 in anesthetized monkey. Since all high-order shape selectivities of neurons emerge from combinations of output of multiple V1 neurons, we calculated response properties to combinations of two local spectral components, each defined at (x, y, fx, fy), where (x, y) is position and (fx, fy) is spatial frequency/orientation. Response magnitude was defined by the product of the amplitudes of the two spectral components. Spectral combinations were examined exhaustively within the sampled frequency domain. We recorded 416 neurons from monkey V2. Half of neurons showed interactions to a combination of two spatially adjacent spectral components, but these response maps were somewhat noisy and Zscores were not so high (<7). On the other hand, many neurons showed the strongest interaction to the combination of two spatially overlapped spectral components with large Z-scores (>7) (N = 149). These results suggest that neurons in V2 code combinations of two spectral components. Research fund: KAKENHI 22135006, 22300110(IO), 18300111 (MI) and Global COE.
1 2
Department of Neuroscience, Columbia University Medical Center, NY, USA Japan Society for the Promotion of Science (JSPS), Tokyo, Japan
The world around us appears to remain stationary even when our eyes are in motion. This perceptual stability is caused by an internal estimation of the external visual world derived from both retinal and extra-retinal signals. However, the human visual system frequently misinterprets the meaning of retinal motion. Karl Duncker (1929) showed that when a stationary object on the retina is embedded in retinal background motion, the subject perceives the object to be moving in a direction opposite from that of the background. When a human or monkey pursues a target moving orthogonally to a flow field, although the eye movement is accurate, the eye movement and target motion direction are perceived to be moving in a diagonal direction away from the flow field. When humans or monkeys make memory-guided saccades to targets flashed at the beginning of an epoch of pursuit across an orthogonal flow field, their eye movements are inaccurate because they compensate for the pursuit movement the subjects thought they made but actually did not. To understand the neuronal mechanisms underlying this illusion we investigated the neuronal responses of pursuit-related neurons in area MST while the monkeys pursued a target moving across an orthogonal flow field and then made a memory-guided saccade to a stimulus which flashed at the beginning of the pursuit epoch. We asked whether the directional selectivity of each pursuit cell in area MST represents the actual direction of the pursuit trajectory or the perceived direction of ongoing pursuit as measured by the error in the memory-guided saccade. We found that the direction tuning profiles of a significant number of MST pursuit cells represented the direction of the perceived pursuit trajectory, not the pursuit eye movement that monkey actually performed. These results suggest that area MST compensates for self-induced motion by using the perceived direction of pursuit, rather than a corollary discharge of the eye movement itself.
doi:10.1016/j.neures.2011.07.1119 doi:10.1016/j.neures.2011.07.1121
P3-j20 From multi-neuron recordings to a brain-based machine: A method to extract and utilize nonlinear population coding manners for constructing artificial visual recognition system Hiroki Kurashige , Hideyuki Cateau RIKEN BSI-TOYOTA Collaboration Center, RIKEN, Wako, Japan We propose a method to construct a brain-based visual recognition system based on the in vivo neural activity data recorded from multiple neurons in the visual cortex of an animal watching visual stimuli. We demonstrate that our method works for a data taken from simulation. Our method extracts visual features represented well by the neural activity of an animal using the so-called kernel canonical correlation analysis (KCCA) which is a multivariate statistical method to detect nonlinearly correlated pairs of visual features and neural activity patterns, and incorporates them into artificial visual recognition systems. The extracted visual features are considered useful for visual recognition because the features are probably used in the brain performing sophisticated visual recognition. We represent the visual features in terms of a kernel function, that we call NC kernel, of kernel method which is good formalism to incorporate the features to artificial recognition systems. Attaching the NC kernel to a support vector machine (SVM), which is frequently used artificial recognition system, we construct a brain-based SVM incorporating the visual features. With the data obtained from the simulation of artificial neural network (ANN) instead of real brain, the “ANN”-based SVM performed the visual discrimination task requiring the knowledge which the ANN has. Our SVM correctly performed the task while usual SVM did not. This result suggests the successful transfer of the visual features from the ANN to the SVM. Moreover, we show that the NC kernel works well even if the visual stimuli presented to the ANN are non-specific to the task for the SVM. This is desired nature because we generally cannot specify the visual stimuli required by the task in the realistic application such as car-mounted camera. We finally discuss the possibility that we can use natural images as task-nonspecific visual stimuli to construct an effective NC kernel in realistic situations. doi:10.1016/j.neures.2011.07.1120
P3-k01 In vivo two-photon imaging analysis of development of direction and orientation selectivity in the mouse visual cortex Madoka Narushima 1 , Nathalie L. Rochefort 2 , Christine Grienberger 2 , Nima Marandi 2 , Arthur Konnerth 2 1
Dept. Physiol., Sch. of Med., Tokyo Women’s Medical Univ., Tokyo, Japan Inst. fuer Neurowissenschaften, Technische Univ. Muenchen, Munich, Germany 2
Previous studies of the ferret visual cortex indicate that the development of direction selectivity requires visual experience. In spite of the usefulness of rodents as a model system for studying mechanisms of functional development of neuronal circuits, it has been less understood that how the rodent visual system develops. Here, we used two-photon calcium imaging to study the development of direction-selectivity in layer 2/3 neurons of the mouse visual cortex in vivo. The number of visual responsive cells gradually increased after eye-opening. Surprisingly, at eye-opening nearly all orientation-selective neurons were also direction-selective. During later development, the total number of neurons responding to drifting gratings increased in parallel with the fraction of neurons that were orientation—but not direction-selective. Our experiments demonstrate that direction-selectivity develops normally in dark-reared mice, indicating that the early development of direction-selectivity is independent of visual experience. Furthermore, remarkable functional similarities exist between the development of direction-selectivity in cortical neurons and the previously reported development of direction-selectivity in the mouse retina. Together, these findings suggest a new experience-independent circuit mechanism for the development of direction-selectivity in the mammalian brain that is distinctly different from that previously found in ferrets. Research fund: JSPS Postdoctoral Fellowship for Research Abroad. doi:10.1016/j.neures.2011.07.1122